Model & Provider
Live Validation

Every API key Gia depends on, probed with a real call and validated for effectiveness. This is the validation gate for the Cost Re-Architecture rollout — no model enters a phase un-tested.

6 GO · 3 BLOCKED

Anthropic, Google, OpenRouter and Brave are GO. OpenAI (GPT-5.5 advisor) is blocked on quota, xAI on a bad key, Perplexity has no key.

Live API key tests

Test = minimal real request expecting a known reply ("PONG") or a valid data payload. Latency is wall-clock for that single call. Run from Gia's container.

Action items

• ✅ Gemini 2.5 Flash / Pro — validated. Cleared to become the new primary + mid-tier.
• ✅ Anthropic Opus 4.8 + Haiku 4.5 — validated. Opus reserved as summon-only advisor.
• ✅ OpenRouter — live, $24.80 remaining. Real balance readable (only provider that exposes true spend).
• ✅ Brave Search — validated. Research path healthy.
• ❌ OpenAI GPT-5.5 — insufficient_quota. BLOCKED as advisor until Byron funds tokens.
• ❌ xAI Grok-4 — invalid API key. Needs a valid key before use.
• ❌ Perplexity — no key configured. Add a key to enable.

This table refreshes whenever keys change (scripts/smoke/probe-keys.sh). Part of the Cost Re-Architecture plan — nothing changed on the live model yet; primary remains Opus 4.8.